Following on the effectiveness of linear adaptive predictors with independently low-order cascaded structures, we investigate a recursively updated lattice implementation of the cascaded forward-backward least mean square (CFBLMS) algorithm. This lattice CFBLMS structure has proven effective in combating the misadjustment and eigenvalue spread effects of the linear prediction process as presented in this paper. Furthermore, experimental results demonstrate that the lattice CFBLMS structure is less affected by quantization distortion. These characteristics translate into better performance in the speed of convergence and the lessening of the misadjustment at bit quantization levels.
展开▼
